Learn · AEO, measured
What makes ChatGPT or Claude name a company in an answer?
Claude names a company mostly when it also cites the company’s own article; GPT-6 names one almost only when asked which vendor or what it costs. That is the pattern in 200 answers we collected on September 28, 2026: 20 buyer questions about AEO services, asked five times each on two engines. Below are the numbers, what they suggest you do, and what they cannot show.
What exactly did we measure?
We asked two AI engines 20 questions a buyer of AEO services might ask, five times each, with web search on: 200 answers in all. No question contained a company name. Claude Opus 5 answered through Claude Code; OpenAI’s GPT-6 Sol answered through OpenAI’s Codex command-line tool. Neither is the consumer app, which can add memory, personalization and its own search settings, so read the results as how the engines behave with search on, not as a screenshot of ChatGPT or claude.ai. A company counted as named when it appeared in the answer text, checked against a list of 94 AEO tools and service providers. ClawWorld sells AEO services and was named in none of the 200 answers.
| Measure | Claude Opus 5 | GPT-6 Sol |
|---|---|---|
| Answers | 100 | 100 |
| Searched the web first | 87 | 99 |
| Named at least one AEO vendor | 93 | 41 |
| Named at least one AEO service provider | 64 | 29 |
| Sources cited per answer, on average | 9.0 | 4.2 |
| Named ClawWorld | 0 | 0 |
Why does Claude name the companies it names?
Claude mostly names a provider in an answer that also cites that provider’s own website. Of the 10 service providers it named most often, 7 were named 107 times between them, and 97 of those answers also cited the provider’s own site. Their cited pages were of four kinds: ranked lists of the best AEO agencies, published by agencies; a guide to what AEO costs; a head-to-head comparison of monitoring tools; and a page written for one specific problem. The last kind was the most direct: one provider’s page about fixing wrong AI answers about a brand got it named in all five answers to our question about an engine mixing a company up with another. We did not check whether each ranked list puts its own author first.
| Claude’s 10 most-named providers | Providers | Times named | Same answer cited their own site |
|---|---|---|---|
| Named with their own pages | 7 | 107 | 97 |
| The other three | 3 | 33 | 9 |
Reputation is the other route. Of the other three, two are long-established agencies that Claude named through other publishers’ lists: their own sites were cited in 2 of the 22 answers that named them. Claude searched the web in 87 of its 100 answers and cited nine sources per answer on average, which leaves room for a company’s own article to be one of them.
When does GPT name a company at all?
GPT-6 Sol named a vendor in all 40 answers to 8 which-provider or what-price questions, and in 1 of 60 answers to the other 12. Those 12 asked how to do something, such as measuring results or changing a website, and for them it cited Google’s and OpenAI’s own documentation rather than naming vendors: developers.google.com appeared in 68 of its 100 answers. It searched the web in 99 of 100 answers, so the missing names are not for lack of searching.
| GPT-6 Sol, by question type | Questions | Answers | Answers naming a vendor |
|---|---|---|---|
| Which provider, or what it costs | 8 | 40 | 40 |
| How to do it | 12 | 60 | 1 |
Where it did name providers, nobody owned the list. GPT-6 Sol named 35 different service providers, 26 of them only once or twice; Claude named 33, only 8 of them once or twice.
So what should a company do to get named?
Treat being cited as the step before being named, and write the pages engines cite for your buyers’ exact questions. On Claude’s side that means pages like the ones above: a comparison, a plain account of cost, a page that answers one specific problem, each opening with a sentence short enough to quote (the writing rules are in how AI answer engines choose what to cite). On GPT-6’s side, the questions worth competing for are the shortlist and price questions. In those answers it introduced small providers by what their own service pages say they do; we did not open every page to confirm this, so treat it as a lead, not a finding. Either way, we expect a company’s pages to show up among the sources before its name shows up in the text, so track both.
What can this data not tell you?
This data is one day, two engines and 20 questions about one category, so read it as a pattern worth testing, not a rule. Two hundred answers show a strong pattern and are too few to rank providers. Every question was about AEO services; a category with different publishers writing about it could behave differently. Both engines ran through developer tools, not the consumer apps. Being cited and being named showed up together, but we did not test whether publishing an article causes an engine to name you. Our list of 94 vendors may miss some of GPT-6 Sol’s long tail, so its counts are a floor. And we sell AEO services, which is a reason to check the method rather than take our word for it: it is written up on our methodology page, and we re-run the same 20 questions from mid-October and publish every round on the ClawWorld Lab.
Is any of this worth paying for?
Paying for AI visibility makes sense once your buyers ask AI before they shortlist, a separate question with its own data. We answer it in is AI visibility worth paying for yet, or is SEO enough?, from 50 baseline reports on seed-stage B2B companies.